Contact us

Giant Swarm Agent Platform

Agents for every team. Control stays with IT.

One open source platform in your own environment. Your experts build the agents, IT keeps approval, cost and the choice of model.

Book a demo Explore on GitHub
  • 100% open source
  • In your environment
  • Built in the EU

Trusted by platform teams at

  • adidas
  • Börse Stuttgart
  • Deutsche Telekom
  • EGGER
  • Vaillant
  • Vodafone
  • WEPA

01 / What a team gets

One feature, spec to production, in a week.

The three agents in the demo above, built by the team that uses them. Each step shows its work and waits for a person where it matters.

  1. 01 Spec

    A specification the team agreed on

    The Specification Assistant asks three questions, then writes goal, scope and acceptance criteria.

    • Under 250 words
    • Ready for the coding agent
  2. 02 Build

    Code and tests, every tool call logged

    The Coding Agent works in a sandbox with the tools IT approved. Every file change and command shows up in the session.

    • Sandboxed
    • Traceable
  3. 03 Ship

    A rollout that waited for one Allow

    The Deployment Assistant summarises the release, asks, and pauses until a person approves. Then region by region.

    • Nothing ships without a person
    • Per region

02 / What IT keeps

Three control points. On every agent, every team.

Approval before anything changes, cost per agent and model, and the choice of model. Set once by IT, applied to every agent anyone builds.

  • Approval

    Nothing is sent until you allow it.

    An agent that wants to change something shows exactly what it will do. A person clicks Allow or Decline.

    Approval The Deployment Assistant waits for a person.
    The approval card in a session: Deployment Assistant wants to run Rollout on Release pipeline, with the arguments and the buttons Decline and Allow.

  • Cost

    Every call tracked.

    Spend per agent, per model and per team, with a budget for each team.

    Usage Spend, tokens and model calls.
    The Usage page: spend, tokens, sessions and active agents, model calls per day and the top agents.

  • Model

    Change the model. Keep the agent.

    Your keys, a subscription or a model you host. The agent, its skills and its data stay where they are.

    Edit agent The model is one card in More options.
    Edit agent, More options: three model cards, ChatGPT subscription, Claude Sonnet and a self-hosted Qwen3.

See it with your own use case.

A 30-minute walkthrough on a lab installation: one of your workflows, the approval and the bill at the end.

Book a 30-minute walkthrough How teams start

03 / Why alone fails

AI adoption stalls when every team solves it alone.

It is an organisation-wide challenge. It cannot be solved tool by tool, team by team.

  1. Security says no.

    Security says yes.

    Agents act with the person's own rights. Every action is traced.

  2. Every team starts from zero.

    Build once, reuse everywhere.

    The next team starts from what the first one built.

  3. Only experts can build.

    Your experts build.

    In plain words, from parts IT has already approved.

  4. The bill is a surprise.

    Cost you can see.

    Per agent, per team and per model, with a budget for each team.

  5. You're locked in.

    Yours to keep.

    Open source, any model, on your own infrastructure.

04 / The platform

One platform. No compromise between business and IT.

The business gets agents where people work. IT keeps identity, approval, guardrails, budgets and visibility.

  • Agent Runtime

    Experts build their own agents.

    Domain experts create and share agents from building blocks IT approved, and reuse what other teams built.

    Agent Runtime Experts build their own agents.

  • Identity and Access

    Agents act with the right permissions.

    An agent acts only with the permissions of the person or role behind it. Every action traced to the agent and the person.

    Identity and Access Agents act with the right permissions.

  • Model Runtime

    Your models, your keys, your bill.

    Frontier providers, European AI clouds or your own open-weight models behind one gateway. Every call attributed to a person and a team.

    Model Runtime Your models, your keys, your bill.

Across all three

Build your own, or start from a curated catalog of models, agents, skills and connectors.

05 / Sovereignty

Sovereignty is having the choice.

In your environment, under your control, with a plug you can pull.

Cost
Every model call visible, with budgets per team. Smaller, focused agents and the right model for each task keep the bill down.
Control
Runs in your environment, on your identity system. Every approval and decision audited.
Accountability
Run by your team or operated by us, by people who run enterprise platforms for a living.
Exit
Open source with no feature gates. If the contract ends, the capability stays with you.
  • 12+ yearsrunning enterprise platforms in production
  • ISO 27001certified information security management
  • Apache 2.0components in the CNCF and the Linux Foundation

06 / Getting started

Start small. Run it your way. Grow from there.

The platform is open source and free to run. You pay for what code alone does not give you: maintainers, signed releases, support and someone accountable for operations.

  1. 01 Start

    The Foundation Package

    A forward deployed engineer sets up the platform in your environment and builds the first agents with you.

    • Fixed price
    • Credited towards operations and support
  2. 02 Run

    Choose who operates it

    Your team runs itWith our releases, maintenance and support behind you.

    We run it for youIn your environment, with someone accountable.

  3. 03 Grow

    Roll it out to every team

    The next team reuses what the first one built. No second setup.

    • Certified catalog
    • Offline channel for air-gapped sites

What you pay for

Get pricing
  • A fixed-price startThe Foundation Package: setup and the first agents, at a fixed price.
  • Operations and supportReleases, maintenance and support, whether your team runs it or we do.
  • Models and computeProvider fees on your own keys, or the compute for models you host.

06 / Questions

Frequently Asked Questions

We already have an AI assistant for our employees. Why this?

Keep it. The Agent Platform runs agents on your own systems and data, in your environment, with your models and your keys.

Can we run the open source version ourselves?

Yes, and we help you start. You pay when it runs in production and you need maintained releases, a patch SLA and someone accountable.

What happens if Giant Swarm disappears?

The whole platform is Apache 2.0 with no feature gates, and its components live in the CNCF and the Linux Foundation. The capability stays with you.

Couldn't our engineers build this themselves?

They could. Then they become AI platform engineers instead of domain experts, and every team repeats the work.

What does it run on?

Wherever you want, as long as there is Kubernetes. If you do not run Kubernetes yet, we provide it.

Read more

Start with one team. Solve it for all of them.

A 30-minute walkthrough on a lab, then a fixed-price start with a forward deployed engineer in your environment.

Book a 30-minute walkthrough Explore on GitHub